The Political Legacy of American Slavery

Avidit Acharya, Stanford University

Matthew Blackwell, Harvard University
Maya Sen, Harvard University

We show that contemporary differences in political attitudes across counties in the American South in part trace their
origins to slavery’s prevalence more than 150 years ago. Whites who currently live in Southern counties that had high
shares of slaves in 1860 are more likely to identify as a Republican, oppose affirmative action, and express racial re-
sentment and colder feelings toward blacks. We show that these results cannot be explained by existing theories, in-
cluding the theory of contemporary racial threat. To explain the results, we offer evidence for a new theory involving
the historical persistence of political attitudes. Following the Civil War, Southern whites faced political and economic
incentives to reinforce existing racist norms and institutions to maintain control over the newly freed African American
population. This amplified local differences in racially conservative political attitudes, which in turn have been passed
down locally across generations.

F or the first 250 years of American history, white land-

owners, predominantly from the South, enslaved mil-
lions of individuals of African descent. This “peculiar
institution,” as it was sometimes called, defined the social,
economic, and political landscape of the American South
throughout this period. Slavery was so crucial to the South
that one Georgia newspaper editor wrote, “Negro slavery is
the South, and the South is negro slavery” (cited in Faust
1988, 60). Yet, despite slavery’s prominence in shaping Amer-
ican history, and despite volumes written by economists and
historians on its consequences, political scientists have largely
overlooked how American slavery and the events following
its abolition could continue to influence the South’s con-
temporary politics. Given recent findings on the long-term
consequences of past events and institutions (Acemoglu,
García-Jimeno, and Robinson 2012; Dell 2010; Nunn and
Wantchekon 2011; Voigtländer and Voth 2012), it would be
surprising if such a fundamental aspect of American history
had no persistent impact on American politics.
In this paper, we show that the local prevalence of
slavery—an institution that was abolished 150 years ago—
has a detectable effect on present-day political attitudes in
the American South. Drawing on a sample of more than
40,000 Southern whites and historical census records, we
show that whites who currently live in counties that had
high concentrations of slaves in 1860 are today on average
more conservative and express colder feelings toward Af-
rican Americans than whites who live elsewhere in the
South. That is, the larger the number of slaves per capita in
his or her county of residence in 1860, the greater the
probability that a white Southerner today will identify as a
Republican, oppose affirmative action, and express attitudes
indicating some level of “racial resentment.” We show that
these differences are robust to accounting for a variety of
factors, including geography and mid-nineteenth-century
economic and social conditions. These results strengthen
when we instrument for the prevalence of slavery using
geographic variation in cotton-growing conditions.
We consider several explanations for our results rooted
in contemporary forces and find each to be inconsistent
with the empirical evidence. For example, we consider the
possibility that whites are simply more racially conservative
when exposed to larger black populations—the central
finding of the literature on racial threat (Blalock 1967;

Avidit Acharya (avidit@stanford.edu) is an assistant professor at Stanford University, Stanford, CA 94305; web: http://stanford.edu/~avidit. Matthew
Blackwell (matt_blackwell@harvard.edu) is an assistant professor at Harvard University, Cambridge, MA 02138; web: http://www.mattblackwell.org. Maya
Sen (maya_sen@hks.harvard.edu) is an assistant professor at Harvard University, Cambridge, MA 02138; web: http://scholar.harvard.edu/msen.
Data and supporting materials necessary to reproduce the numerical results in the paper are available in the JOP Dataverse (https://dataverse.harvard.edu
/dataverse/jop). An online appendix with supplementary material is available at http://dx.doi.org/10.1086/686631.

The Journal of Politics, volume 78, number 3. Published online May 19, 2016. http://dx.doi.org/10.1086/686631
q 2016 by the Southern Political Science Association. All rights reserved. 0022-3816/2016/7803-0001$10.00 621

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
622 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
Blumer 1958; Key 1949). However, when we estimate the
direct effect of slavery on contemporary attitudes (Acharya,
Blackwell, and Sen, forthcoming), we find that contempo-
rary shares of the black population explain little of slavery’s
effects. We also test various other explanations, including
the possibility that slavery’s effects are driven exclusively by
twentieth-century population shifts or income inequality
between African Americans and whites. We find no evi-
dence that these contemporary factors and theories of pop-
ulation sorting fully account for our results. Introducing
individual-level and contextual covariates commonly used in
the public opinion literature also does not explain away our
finding.
To explain our results, we instead propose a theory of
the historical persistence of political attitudes. The evidence
suggests that regional differences in contemporary white at-
titudes in part trace their origins to the late slave period and
the time period after its collapse, with prior work suggest-
ing that the fall of slavery was a cataclysmic event that un-
dermined Southern whites’ political and economic power.
For example, Du Bois (1935), Foner (2011), and Key (1949),
among others, have argued that the sudden enfranchisement
of blacks was politically threatening to whites, who for cen-
turies had enjoyed exclusive political power. In addition, the
emancipation of Southern slaves undermined whites’ eco-
nomic power by abruptly increasing black wages, raising la-
bor costs, and threatening the viability of the Southern plan-
tation economy (Alston and Ferrie 1993; Ransom and Sutch
2001). Taken in tandem with massive preexisting racial hos-
tility throughout the South, these political and economic
changes gave Southern Black Belt elites an incentive to fur-
ther promote existing anti-black sentiment in their local com-
munities by encouraging violence toward blacks and racist
attitudes and policies (Roithmayr 2010). This amplified the
differences in white racial hostility between former slave-
holding areas and non-slaveholding areas and intensified
racially conservative political attitudes within the Black Belt.
These have been passed down locally, one generation to the
next.
We provide empirical support for this mechanism by
showing that areas of the South that were the earliest to
eliminate the political and economic incentives for anti-black
violence—for example, by adopting new technologies, such
as tractors, that reduced the demand for black farm labor—
are also the areas in which slavery’s long-term effects have
most attenuated. Furthermore, as evidence for intergenera-
tional (cultural) transfer of attitudes, we show that there exists
a strong correlation between the racial attitudes of parents and
their children in the US South. Our evidence, therefore, sup-
ports the theory that political attitudes have persisted histor-
ically in the US South, rather than the view that attitudes
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
are driven exclusively by contemporaneous forces—making
our position quite distinct from much of the existing public
opinion literature.
The paper proceeds as follows. First, we motivate our
hypothesis that historical persistence—and not just con-
temporary factors—shape modern-day political attitudes.
We discuss our data in the next section and present our
core results linking the prevalence of slavery in 1860 and
contemporary attitudes in the following section, with ad-
ditional robustness checks presented in the appendix, avail-
able online. We then consider and provide evidence against
several competing theories rooted in contemporary factors,
including the theory of racial threat. In the following section,
we provide evidence for our theory of the historical persis-
tence of political attitudes, paying close attention to post-
bellum political and economic incentives as the driving mech-
anism. We conclude by discussing the broader implications
of our research for scholarship in American political behavior.
EXPLAINING REGIONAL DIFFERENCES IN
SOUTHERN POLITICAL AND RACIAL ATTITUDES
We orient our analysis toward the Southern “Black Belt”
(or the “Cotton Belt”), the hook-shaped swath of land that
was the primary locus of antebellum slavery (fig. 1). Scholars
have noted that Black Belt whites were particularly promi-
nent in Southern politics and have been more conservative
than whites elsewhere in the South. As V. O. Key wrote, it
is “the whites of the black belts who have the deepest and
most immediate concern about the maintenance of white
supremacy,” and “if the politics of the South revolves around
any single theme, it is that of the role of the black belts” (Key
1949, 5–6). Furthermore, the Black Belt has had an enormous
influence on national politics. Members of Congress from
these areas held influential positions, effectively exercising
veto power during the development of the welfare state in
the 1920s and 1930s (Katznelson, Geiger, and Kryder 1993).
Given these facts, our motivating question is this: Why are
whites who currently live in the Black Belt more conserva-
tive than whites living elsewhere in the South, particularly
on race-related issues? We consider two broad classes of
explanations: (i) the historical persistence of attitudes orig-
inating in slavery and (ii) contemporary factors, including
contemporary demographics and geographic mobility.
Historical persistence of white political attitudes
Our first hypothesis is that today’s Black Belt is more po-
litically conservative than other parts of the South in part
because of its history of chattel slavery. We are motivated in
this hypothesis by an emerging empirical literature showing
that the effects of coercive institutions persist in other

Figure 1. Estimated proportion slave in 1860 by county
contexts. Dell (2010), for example, shows that a colonial
forced labor system in Peru and Bolivia led to lower levels
of modern-day household consumption and childhood
growth. Acemoglu et al. (2012) find that the use of slaves in
the colonial gold mines of Colombia predicts modern-day
poverty, reduced school enrollment, and decreased vacci-
nation rates. Nunn and Wantchekon (2011) show that Af-
ricans whose ancestors were targeted by the slave trade
have higher levels of mistrust today than other Africans.
Within the United States, O’Connell (2012) demonstrates
that areas of the American South that had high numbers of
slaves have greater economic inequality between blacks and
whites today. Similarly, Lagerlöf (2005) and Nunn (2008)
find a negative relationship between the prevalence of slavery
and income in the American South, and Mitchener and
McLean (2003) find a negative relationship between slavery
and modern-day labor productivity. These papers are part of
a growing literature that shows that historical institutions
such as slavery can affect both institutional and behavioral
outcomes long after the institutions themselves disappear
(Nunn 2009). This work complements an existing literature
documenting the path dependence of historical institutions
over time (e.g., Pierson 2000).
Building from this literature, we hypothesize that South-
ern slavery may have had a similarly lasting effect on political
and racial attitudes. The rise and swift fall of chattel slavery
together were cataclysmic events. Specifically, the eventual fall
of slavery undermined the political and economic power of
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Volume 78 Number 3 July 2016 / 623
the Southern whites, particularly in the Black Belt (Du Bois
1935; Foner 2011), making them more hostile toward African
Americans and conservative in their political, racial, and
economic views (noted contemporaneously by Key 1949).
Qualitative accounts (as we document below) suggest that
the nature of Southern whites’ responses to the collapse of
slavery varied according to how locally prevalent—and thus
politically and economically important—slavery had been.
Areas with more enslaved people reacted more sharply to
emancipation by curtailing blacks’ rights and oppressing newly
freedmen and their mobility.
In addition, a large literature has shown that attitudes
can persist historically through both cultural and institutional
channels (Nunn and Wantchekon 2011; Voigtländer and Voth
2012). On the one hand, Southern institutions such as Jim
Crow helped enforce racial segregation, while racially targeted
violence reinforced practices of black subjugation (Woodward
[1955] 2002). On the other hand, the culture of the Southern
Black Belt was one where black subjugation was passed on
within white families and across generations—a process that
no doubt included intergenerational socialization (Bisin and
Verdier 2000; Boyd and Richerson 1988; Campbell et al. 1980;
Jennings and Niemi 1968).
Based on these arguments, we expect that (i) areas that
were more reliant on slavery should be more conservative
today on race-related issues and in terms of party affilia-
tion, (ii) race-related attitudes should be correlated across
generations within the South, and (iii) the effects of slavery
624 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
should be weaker (i.e., should have decayed more) in areas
where the incentives for anti-black attitudes faded earlier.
How contemporary demographics could explain
regional variation in white political attitudes
In contrast to the arguments above, much of the political
science literature points to contemporary (not historical)
forces as providing the explanation for why Black Belt whites
are more conservative on race. By and large, the literature has
interpreted Key’s (1949) work as suggesting that whites con-
temporaneously become more conservative when they are ex-
posed to the high concentrations of African Americans who
live in their communities.1 The high concentration of African
Americans in today’s Black Belt could contemporaneously
threaten white dominance, resulting in whites actively choos-
ing more conservative political beliefs today. The literature
supporting this idea, known as “racial threat,” is voluminous.2
For example, Glaser (1994) finds evidence linking negative
white attitudes toward civil rights or African American pol-
iticians with high concentrations of blacks. Giles and Buckner
(1993) find a relationship between black concentrations and
white support for racially conservative candidates such as
David Duke (these findings are, however, challenged by Voss
[1996]). This literature, however, has not considered that
slavery could be an independent predictor of contemporary
attitudes (apart from its effect on contemporary demograph-
ics), making it an omitted variable in studies of racial threat
in the South.
Other aspects of the contemporary local context may also
affect white attitudes—for example, income gaps between
blacks and whites, urban-rural differences, and other con-
textual and individual-level factors (e.g., Hopkins 2010;
Oliver and Mendelberg 2000).3 A final category of explana-
tions concerns white mobility through the twentieth century.
For example, it could be that more racially conservative
whites have migrated into former slaveholding areas, while
racial liberals have left, thereby creating a regional pattern in
1. As we note below, Southern slavery is correlated with contemporary
black concentration, making it difficult to disentangle the effects of slavery
from the effects of contemporary black concentrations.
2. Early studies showed, e.g., that modern black concentrations pre-
dict white support for segregationist candidates such as George Wallace
(e.g., Wright 1977), racially hostile white attitudes (Blalock 1967; Giles
1977), negative attitudes on school desegregation (Ogburn and Grigg
1956), and higher incidence of lynchings (Reed 1972).
3. Some work has even highlighted the connection between slavery and
these contemporary factors (Mitchener and McLean 2003; Nunn 2008;
O’Connell 2012). While these papers suggest that slavery might affect con-
temporary attitudes indirectly through contemporary factors such as eco-
nomic inequality and prosperity, we find below that slavery has a direct effect
on contemporary attitudes that does not work through these channels.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
political attitudes that is less about persistence of beliefs and
more about the sorting of beliefs.
Ultimately much of the public opinion literature focuses
on contemporary or individual-level factors in explaining
political beliefs rather than on historical forces. Yet Key him-
self was aware of the importance of history in the context of
slavery when he noted that, in the years leading to the Civil
War, “those with most at stake—the owners of large numbers
of slaves—were to be found roughly in the same areas as
present-day black belts” (Key 1949, 6). We now turn to ex-
ploring this historical link in terms of regional variation in
Southern whites’ attitudes.
HISTORICAL SLAVE DATA AND CONTEMPORARY
PUBLIC OPINION DATA
Our main explanatory variable and proxy for slavery’s
prevalence is the proportion of each county’s 1860 popula-
tion that was enslaved, as measured by the 1860 US Census.
Although counts of enslaved people were taken before 1860,
we use measures from 1860 because they represent the last
record before chattel slavery was abolished in 1865. In ad-
dition, white planters were very mobile in the antebellum
period, during which slaves (not land) were their main source
of wealth; after emancipation, mobility decreased rapidly as
white elites became increasingly oriented toward landowning
(Wright 1986, 34). If any local legacy exists, we would expect
to see it in data from 1860. Since county boundaries have
shifted since 1860, we use an area-weighting method to map
data from the 1860 Census onto county boundaries in 2000,
enabling us to estimate the proportion enslaved in 1860
within modern-day counties.4 Figure 1 depicts the data. Over-
all, we have in our data approximately 4 million enslaved
people, constituting 32% of the Southern population.
Outcome variables measuring contemporary white
political and racial attitudes
We analyze three county-level outcome measures, which
come from the Cooperative Congressional Election Study
(CCES), a large survey of American adults (Ansolabehere
2010). We pool CCES data from the 2006, 2008, 2009, 2010,
and 2011 surveys to create a combined data set of over
4. Total population and total enslaved population in 1860 counties are
divided among the counties in 2000 so that the proportion of the 1860
population from 1860-county i that is assigned to 2000-county j is based
on the size of their overlapping areas. This approach produces estimates
and results similar to those provided by (i) O’Connell (2012) (r p 0.986),
who uses an alternative interpolation technique, and (ii) a simpler method
that relies on matching counties by name. See appendix section A for
more information on our approach.

157,000 respondents. We subset these data to the former
Confederate states plus Missouri and Kentucky, both of
which had significant internal support for the Confederacy,5
and to self-identified whites, leaving us with more than
40,000 respondents across 1,329 of the 1,435 Southern coun-
ties. In addition, we also investigate individual-level black-white
thermometer scores from waves of the American National
Election Survey (ANES) from 1984 until 1998, a time period
in which the ANES both used a consistent sampling frame
and included county-level identifiers for respondents. After
restricting the sample to Southern whites, we have an ANES
sample of 3,123 individuals across 64 counties in the South.
This makes the ANES more restricted in its geographic cov-
erage, but it contains valuable direct questions on the sub-
jective evaluation of racial groups.
The four outcome measures are as follows.
Partisanship. We examine partisanship, because, as many
scholars have argued, Southern whites’ partisanship (and
partisan re-alignment) has been intimately connected to, and
reflective of, their attitudes on race and black-white relations
(Carmines and Stimson 1989; Key 1949; Kuziemko and Wash-
ington 2015; Valentino and Sears 2005). Such partisan iden-
tification can not only reflect racial attitudes, as suggested
by these papers, but may also reflect beliefs on policy issues
closely related to race, including redistribution (Gilens 2009;
Lee and Roemer 2006). Partisanship also serves as an im-
portant bridge to regional and national politics.
We construct our partisanship measure from a standard
seven-point party identification question on the CCES. We
operationalize the party variable as whether an individual
identified at all with the Democratic Party (1 if Democrat;
0 otherwise).6 Thus, the county-level measure represents
the proportion of whites in each county who identified as
Democrats.
Support for affirmative action. All CCES surveys ask re-
spondents whether they support or oppose affirmative action
policies, which are described as “programs [that] give pref-
erence to racial minorities and to women in employment and
college admissions in order to correct for discrimination”
5. The sample thus includes Alabama, Arizona, Arkansas, Florida,
Georgia, Kentucky, Louisiana, Mississippi, Missouri, North Caroline,
South Carolina, Tennessee, Texas, Virginia, and West Virginia.
6. We use survey data as opposed to voter registration data because
primaries in many Southern states are open. Coupled with the dramatic
changes in partisanship in the South over the last 40 years, this means voter
registration data are unreliable measures of current partisan leanings. Fi-
nally, survey data allow us to focus on the partisanship of whites voters only.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Volume 78 Number 3 July 2016 / 625
(2008 CCES).7 We construct the outcome variable by col-
lapsing the four-point scale, from “strongly support” to
“strongly oppose,” to an indicator representing whether the
respondent demonstrated any level of support for affirma-
tive action (1 for support; 0 otherwise). At the county level,
then, this is the proportion of whites who say that they
support affirmative action.
Racial resentment. Kinder and Sears (1981, 416) write that
racial resentment (or symbolic racism) “represents a form of
resistance to change in the racial status quo based on moral
feelings that blacks violate such traditional American values
as individualism and self-reliance, the work ethic, obedience,
and discipline.” We construct a third outcome variable us-
ing the two CCES questions on racial resentment. The first
question, asked in the 2010 and 2011 CCES surveys, asks
respondents on a five-point scale whether they agree with
the following statement: “The Irish, Italian, Jews and many
other minorities overcame prejudice and worked their way
up. Blacks should do the same.” The second question, asked
in 2010, asks respondents, also on a five-point scale, whether
they agree that “Generations of slavery and discrimination
have created conditions that make it difficult for Blacks to
work their way out of the lower class.” For the 2010 CCES,
when both questions were asked, we rescaled both questions
and averaged them to create one measure. The final county-
level measure is the average level of agreement with the
racially resentment statement on a five-point scale.
White-black thermometer difference. In many years, the
ANES contains “feeling thermometer” questions, which ask
respondents to evaluate their feelings about politicians and
groups (including racial or ethnic groups) on a scale from 0
to 100.8 Since these scales have engendered criticisms that
the ratings fluctuate heavily from individual to individual
(Wilcox, Sigelman, and Cook 1989) and that they are less
stable than party identification (Markus and Converse
7. Although the question wording differs across years, we have no
reason to believe that these wording variations affect our analysis.
8. The 1984 ANES gave respondents the following instructions:
I’ll read the name of a person and I’d like you to rate the person
using the feeling thermometer. Ratings between 50 degrees and
100 degrees mean that you feel favorable and warm toward the
person. Ratings between 0 degrees and 50 degrees mean that you
don’t feel favorable toward the person and that you don’t care too
much for that person. You would rate the person at the 50 degree
mark if you don’t feel particularly warm or cold toward that
person.
For groups like “blacks” or “whites,” the instructions asked “And, still
using the thermometer, how would you rate the following?”
626 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
1979), we use them mainly to buttress the findings from our
previous three outcome variables. To measure relative racial
hostility, we take the difference between white respondents’
feeling thermometer ratings toward whites and their feeling
thermometer ratings toward blacks. Thus, a positive differ-
ence would indicate that respondents have warmer feelings
toward whites as opposed to blacks. Only using black ther-
mometer scores yields similar results, but we use the differ-
ence in case slavery has an overall effect on racial group
thermometer ratings.
Appendix tables A.1 and A.2 report summary statistics
for these and other data. Figure 2 depicts the bivariate re-
lationships between proportion slave in 1860 and the four
outcome measures from the CCES and ANES. It shows neg-
ative, statistically significant relationships between slave prev-
alence and proportion Democrat and support for affirmative
action and positive, statistically significant relationships with
racial resentment and thermometer score differences. We now
turn to establishing plausible estimates of the causal effect of
slavery’s prevalence on these outcomes.
SLAVERY’S EFFECTS ON CONTEMPORARY
OUTCOMES
In models that analyze the CCES outcomes, we conduct a
county-level analysis using weighted least squares (WLS)
with the within-county sample size (appropriately weighted
by the sampling weights) as weights.9 We opt for analyses at
the county level for two reasons. First, the key variable of
interest here (in effect, our treatment) is slavery in 1860,
which is only measured at the county level, meaning we
cannot detect any within-county variation in this measure.
Second, when attempting to estimate the effect of a cluster-
level variable such as this, Green and Vavreck (2008) show
that aggregating units up to the cluster level provides the
most accurate variance estimates. While it is sometimes use-
ful to trade off this accuracy on standard error estimation to
allow for the inclusion of individual-level covariates, in our
context many individual-level covariates are potentially di-
rectly affected by the treatment (slavery), so including them
may introduce posttreatment bias (Rosenbaum 1984). In spite
of this, we present respondent-level analyses in appendix
table A.3 that include additional respondent-level controls
and standard errors clustered at the county level. These re-
sults are consistent with our county-level results below, al-
though we caution that the estimates may be biased. Fur-
9. Whites in high-black areas are less likely to be sampled than whites
in low-black areas, since the sampling reflects the population distribution
of whites. If we were making inferences about both blacks and whites, this
approach would be problematic. However, we use the white population
subset to make inferences only about the Southern white population.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
thermore, we opt for an individual-level analysis for our
results on NES thermometer scores since we are collecting
together 15 years’ worth of survey results wherein racial
attitudes have changed dramatically, making it important to
include survey-year fixed effects in our models. The level of
analysis has very little impact on the substantive results, and
the ANES results at the county level are presented in ap-
pendix table A.11.
In table 1, we report the baseline estimates of slavery’s
effect on the three CCES outcomes. Column 1 of table 1
presents the simple WLS relationship between slavery and
white partisan identification, which measures the relation-
ship depicted in figure 2. In the remaining columns, we in-
clude state-level fixed effects to address the possibility that
states adopted different policies that could have influenced
slave shares in 1860 and could affect our outcome variables
in ways unrelated to slavery. In addition, in the remaining
columns, we control for factors that may have been predic-
tive of proportion slave in 1860. These “1860 covariates,”
unless otherwise noted, come from the 1860 US Census and
address possible differences between slaveholding and non-
slaveholding counties. First, since wealthier or more popu-
lous counties may have had more or fewer slaves, we control
for economic and demographic indicators from 1860. These
include (i) the log of the total county population, (ii) the
proportion of farms in the county smaller than 50 acres,
(iii) the inequality of farmland holdings as measured by the
Gini coefficient for landownership (Nunn 2008), (iv) the log
of total farm value per improved acre of farmland in the
county, and (v) the log of the acres of improved farmland.
Second, because counties may have had different norms about
race, we include controls for (vi) the proportion of total pop-
ulation in 1860 that is free black. We also control for char-
acteristics related to trade and commerce, including separate
indicators for whether the county had access to (vii) rails and
(viii) steamboat-navigable rivers or canals. Finally, to account
for any remaining spatial variation, we control for (ix) the log
of the county acreage, (x) the ruggedness of the county terrain
(Hornbeck and Naidu 2014), and (xi) the latitude and lon-
gitude of the county, as well as their squared terms (to flexibly
control for spatial variation in the outcome). For all of these
variables, we mapped 1860 data onto modern county
boundaries using the procedure described in appendix sec-
tion A.10
10. In other results, we have additionally controlled for antebellum
county-level vote share for the Democrats in various presidential elections.
These controls do not affect the results but have the unfortunate by-
product of dropping South Carolina, which decided its electoral votes by
the state legislature. For that reason, we do not include this particular
control here.

Figure 2. Bivariate relationships between proportion slave in 1860 and the four outcome measures with a linear fit. All four relationships are significant at
p ! .05 significance levels. Size of the points are in proportion to their within-county sample size (weighted by sampling weights).
As columns 2–4 show, the conditional effects of slavery
are meaningful and significant for all three CCES outcome
variables. To illustrate, a 20 percentage point increase in the
slave proportion (roughly a one standard deviation change)
is associated with a 2.3 percentage point decrease in the
share of whites who currently identify as Democrats (and
so roughly a 4.6 percentage point shift toward the Repub-
lican Party), a 3 percentage point decrease among those who
currently support affirmative action, a 0.08 point increase in
the average racial resentment score in the county, and a 7 point
increase in the county average thermometer preference for
whites relative to blacks. These represent approximately a
0.09–0.13 standard deviation change in the CCES outcomes
and roughly a 0.32 standard deviation change in the ther-
mometer score difference. The individual-level results for
the CCES in appendix table A.3 give very similar estimates,
making it unlikely that these results are due to some sort of
ecological or aggregation bias.11 Finally, we present results
for each racial thermometer score separately in appendix
table A.10 and the full set of county-level results for ther-
mometer scores in appendix table A.11, both of which are
consistent with results in table 1.
Instrumenting for slavery with cotton suitability
There are two potential concerns with the above as a causal
analysis. First, the 1860 slave data are historical and may be
measured with error. Second, we may have inadequately
controlled for all pre-1860 covariates that simultaneously
11. In appendix table A.4, we present results with a full set of inter-
actions between slavery and individual-level covariates, and we find no
significant interactions with education, household income, religiosity,
gender, or age. In the same table, we show that there are no consistent
interactions with contemporary contextual variables as well.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Volume 78 Number 3 July 2016 / 627
affect slave proportion in 1860 and political attitudes today,
which might result in a spurious relationship. To allay some
of these concerns, we present additional robustness checks,
matching analyses, and counterfactual comparisons be-
tween the North and the South in the appendix. Here, we
present results from an alternative specification that instru-
ments for slave proportion in 1860 with county-level mea-
sures of the environmental suitability for growing cotton.
We constructed these measures using data from the UN
Food and Agriculture Organization (FAO).12 Table 2 pres-
ents our instrumental variable (IV) estimates of the effects
of proportion slave on the four outcome measures using a
two-stage least squares (2SLS) model with state fixed effects,
log of the county size, ruggedness of the terrain, water ac-
cess, latitude and longitude, and their squared terms in-
cluded as controls in both stages.13 Column 1 presents the
first-stage relationship between cotton suitability and pro-
portion slave. Columns 2–5 present the second-stage estimates
of the effect of proportion slave on the outcome measures.
12. These measures represent the maximum potential cotton yield
based on soil, climate, and growing conditions. The estimates are based on
climate averages from 1961 to 1990, and we average the “intermediate”
and “high” levels of inputs, which refer to the effort required to extract the
resource. We omit suitability for other crops, such as tobacco, because
they have no relationship with slavery conditional on cotton suitability.
While these measures use data from the contemporary period, we expect
most of the changes to the suitability between 1860 and 1960 to either be
uniform shifts across the entire region due to worldwide climate change or
be unrelated to local political attitudes.
13. To strengthen the internal validity of our design and minimize the
potential for confounding, we omit counties with zero cotton suitability
from the analyses. We use county-level data for the ANES measures to
side-step the issue of clustering by county in the 2SLS model.
628 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen

Table 1. Effects of Slavery on White Political Attitudes

Support for Racial White-Black

Proportion Democrat Affirmative Action Resentment Thermometer Difference

(1) (2) (3) (4) (5)

Proproportion slave, 1860 2.217** 2.127** 2.145** .463** 36.125**

(.023) (.043) (.036) (.139) (9.211)
Level County County County County Individual
1860 covariates √ √ √
State fixed effects √ √ √ √
State-year fixed effects √
Clustered SEs √
N 1,242 1,152 1,152 1,027 1,489
R2 .065 .203 .101 .143 .183

Note. Models 1–4 are weighted least squares at the county level, with within-county sample sizes, adjusted by sampling weights, as weights. Model 5 is
weighted least squares at the individual level with state-year fixed effects. Standard errors, in parentheses, are clustered at the county level in model 5. There
are 50 counties in model 5.
** p ! .01.

The results show second-stage estimates that are stronger
than our baseline estimates, reported in table 1.
For the IV approach to serve as a plausible identification
strategy, cotton suitability must have an effect on contem-
porary attitudes exclusively through slavery, a strong as-
sumption. Cotton suitability could, for example, determine
how rural a county is today, which in turn could affect po-
litical attitudes. While the exclusion restriction is an un-
testable assumption, we assess its plausibility using a falsi-
fication test motivated by Nunn and Wantchekon (2011).
We first estimate the reduced-form relationship between
cotton suitability and contemporary beliefs both within and
outside of the South—that is, mostly in the North. The legal
absence of slavery in the North in this time period means that
cotton suitability cannot affect political attitudes through slave
prevalence. Any relationship between cotton suitability and
political attitudes in the North would be a direct effect of
cotton suitability on political attitudes. In appendix section B.2,
we present results from this analysis showing that, outside the
South, the relationship between cotton suitability and political
attitudes is either very small or in the opposite direction as
in the South. This lends some additional credibility to our IV
approach.
EXPLANATIONS FOR SLAVERY’S EFFECT ON
CONTEMPORARY WHITE ATTITUDES BASED ON
EXISTING THEORIES OF POLITICAL BEHAVIOR
We now turn to explaining the above findings. We first
consider three possible explanations rooted in contempo-
rary factors, as opposed to historical ones: racial threat,
geographic sorting, and contemporary income inequality
between blacks and whites, which could lead to statistical
discrimination. (We examine two other explanations—ru-
ral/urban differences and the effects of Civil War destruc-
tion—in the appendix.) After finding no support for these
contemporary factors, we turn in the following section to
what we believe is the more likely explanation concerning
the long-term historical persistence of political attitudes.
Racial threat (contemporary black concentrations)
As noted above, one plausible explanation for our results is
that they are driven by contemporary black concentrations,
which correlate highly with slavery. Indeed, the correlation
between percent slave in 1860 and percent black in 2000 is
0.77. Since the local prevalence of slavery has produced
high concentrations of blacks in the modern-day Black Belt,
whites living in the Black Belt could be more conservative
today simply due to modern racial threat.
To assess whether modern racial threat accounts for our
findings, we check how much of our baseline results can be
explained by contemporary black concentrations. We do so
in two ways. First, we include the “mediator” (here, pro-
portion black in 2000 as measured by the 2000 US Census)
as a covariate in the baseline specification, along with the
treatment of interest (proportion slave in 1860). This anal-
ysis is shown in table 3, columns 1, 3, and 5. The coefficient
on proportion slave in 1860 remains significant and actually
strengthens, suggesting that its direct effect does not operate
through proportion black in 2000.

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Volume 78 Number 3 July 2016 / 629

Table 2. Instrumental Variables Estimates of the Effect of Slavery

Proportion Proportion Support for Racial White-Black

Slave, 1860 Democrat Affirmative Action Resentment Thermometer Difference
(1) (2) (3) (4) (5)

Cotton suitability .441**

(.037)
Proportion slave, 1860 2.277** 2.247** .823* 51.748*
(.104) (.088) (.352) (19.305)
State fixed effects √ √ √ √ √
Geographic controls √ √ √ √ √
Model 2SLS 2SLS 2SLS 2SLS 2SLS
First stage Second stage Second stage Second stage Second stage
N 1,120 1,120 1,120 998 51
F-statistic 80.077** (df p 21; 1,098)

Note. Model 1 is the first-stage relationship. Models 2–5 are weighted two-stage least-squares models, with weights as given in table 1. Standard errors are in
parentheses.
* p ! .05.
** p ! .01.

These estimates, however, could suffer from posttreat-
ment bias (Rosenbaum 1984). After all, the modern geo-
graphic distribution of blacks is a direct consequence of the
prevalence of slavery (as noted by Key [1949]). In addition,
including the proportion black today as a mediator in a
mediation approach would violate the key assumption of
no intermediate confounders (Imai et al. 2011). We address
these twin concerns by using a method developed in bio-
statistics by Vansteelandt (2009) and applied to other so-
cial science contexts by Acharya et al. (forthcoming). This
method enables us to calculate the controlled direct effect
of slavery, which is the effect of slavery on our outcomes if
we were to fix the modern-day concentration of African
Americans at a particular level (see Acharya et al. [forth-
coming] for a technical overview of the methodology). To
implement the method, we use a two-stage estimator, called
the sequential g-estimator, that estimates a controlled di-
rect effect of slavery when we have a set of covariates that
satisfies the assumption that there exist no omitted vari-
ables for two relationships: one between the outcome and

Table 3. Effects of Slavery on White Attitudes Net the Effect of the Contemporary Proportions of African Americans

Proportion Democrat Affirmative Action Racial Resentment

(1) (2) (3) (4) (5) (6)

Proportion slave, direct effect 2.150** 2.130* 2.153** 2.130** .514** .472*
(.043) (.057) (.037) (.042) (.140) (.184)
Proportion black, 2000 .172** .062 2.416**
(.043) (.037) (.140)
State fixed effects √ √ √ √ √ √
1860 covariates √ √ √ √ √ √
Bootstrapped SEs √ √ √
Model WLS Seq. g-est. WLS Seq. g-est. WLS Seq. g-est.
N 1,152 1,152 1,152 1,152 1,027 1,027
R2 .214 .209 .103 .095 .151 .144

Note. Columns 1, 3, and 5 simply include proportion black in the year 2000 as an additional control to the baseline specification from table 1. Columns 2, 4,
and 6 use sequential g-estimation (Seq, g-est) of Vansteelandt (2009). WLS p weighted least squares. Standard errors are in parentheses.
* p ! .05.
** p ! .01.

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
630 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
proportion slave in 1860 and the other between the out-
come and proportion black in 2000 (Acharya et al., forth-
coming; Vansteelandt 2009).14 To estimate the direct effect
of slavery, we first estimate the effect of contemporary black
concentrations on whites’ views today, controlling for all of
our 1860s covariates and including the additional interme-
diate covariates in footnote 15.15 We then transform each of
the outcome variables by subtracting the effect of black con-
centrations to create counterfactual estimates of the outcomes
as if all counties had the same proportion black today.16 Fi-
nally, we estimate the effect of proportion slave on this trans-
formed variable, which gives us the controlled direct effect of
proportion slave on our measures of contemporary political
attitudes.
Estimates from this analysis are reported in columns 2,
4, and 6 of table 3.17 Compared to the baseline estimates of
table 1 and the potentially biased estimates in columns 1, 3,
and 5, these results demonstrate that contemporary black
shares have little influence on slavery’s effect on any of the
outcomes. The direct effects of slave proportion are similar
to those in table 1 and are still highly significant. Moreover,
once we account for slavery in 1860, contemporary black
concentrations appear to have the opposite effect that racial
threat theory would predict for Southern white attitudes.
Finally, with the full controls from the first stage of the
sequential g-estimator, the effect of proportion black today
is no longer significant (appendix table A.12). Thus, we see
no evidence that slavery’s effects operate via contemporary
black concentrations.
Geographic sorting
The next possibility is that population sorting explains our
results. For example, racially hostile whites from other parts
of the South (or elsewhere) may have migrated to former
slave counties during the last 150 years. Analogously, whites
14. That is, we want to control for intermediate factors that possibly
led from slavery in 1860 to black concentrations today (and that are
correlated with white attitudes today). To determine this set of covariates,
we draw on the racial threat literature (see, e.g., Giles and Buckner 1993)
and include log population in 2000, unemployment in 2000, percent of
individuals with high school degrees in 1990, and log median income in
2000. These results assume no interaction between proportion slave and
contemporary proportion black, but weakening this assumption does not
change the findings (Acharya, forthcoming).
15. Results from this first stage can be found in appendix table A.12.
16. That is, we take the coefficient on proportion black, multiply that
by the actual share black in the county, and then subtract this product
from each of the three white attitudinal outcome variables.
17. To account for the added uncertainty of the two-step nature of
sequential g-estimation, we report bootstrapped standard errors.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
who hold more racially tolerant beliefs may have continually
left former slaveholding areas.
For geographic sorting to explain our results, patterns of
mobility into (and out of ) the former slaveholding areas
would have to differ from non-slaveholding areas. To in-
vestigate this possibility, we look into patterns of migration
in a five-year snapshot from 1935–1940, drawing on the
public use microsample (PUMS) of the 1940 US Census
(Ruggles et al. 2010). This year of the census is unique in
that it provides the county in which a person resided in
1935 and in 1939. These data allow us to investigate if white
migrants into or out of former slave areas during this time
period were somehow distinct from other white migrants.
If sorting plays an important role in our results, we would
expect to see differences between migrants to/from high-
slave areas versus low-slave areas. To test for differences
among out-migrants, we adopted the following strategy:
we ran a regression of individual characteristics on out-
migration status for white respondents, the proportion of
slaves in the respondent’s 1935 county of residence, and the
interaction between the two. We also included the 1860
covariates and state fixed effects for the 1935 counties. The
interaction in this regression measures the degree to which
differences between out-migrants and those who didn’t
migrate varies as a function of proportion slave. For in-
migration, we take a similar approach but replace the char-
acteristics of the 1935 county of residence with the char-
acteristics of the 1939 county of residence.
Figure 3 depicts the results from this analysis and shows
the differences between white migrants and nonmigrants
across a number of characteristics.18 The figure depicts how
these effects vary by proportion slave. For continuous out-
comes, the effects are in terms of standard deviations; for
binary outcomes, they are in terms of differences in pro-
portions. With age, for example, this figure shows that both
out-migrants and in-migrants are significantly younger
compared to those who remained in their counties. How-
ever, the key point is that this pattern is the same for low-
slave and high-slave counties, as evidenced by the over-
lapping estimates. Thus, migrants into or out of high-slave
counties are not any younger than migrants from or to
other counties. This pattern holds more generally: migrants
in the 1935–40 period are distinct from nonmigrants, but
those differences are fairly constant across proportion slave
in the county. Indeed, across all of these characteristics,
there are no significant interactions between migrant status
and proportion slave, meaning that migrants from/to high-
18. See the IPUMS documentation at https://usa.ipums.org/ for a
complete description of these census measures.
 Volume 78 Number 3 July 2016 / 631

Figure 3. Characteristics of white out-migrants and in-migrants compared to white nonmigrants for high-slave and low-slave counties, where migration took
place between 1935 and 1939. In the left panel, each point is the estimated difference between nonmigrants and out-migrants from high-slave areas (dots)
and between nonmigrants and out-migrants from low-slave areas (triangles), conditional on 1860 covariates of the individual’s 1935 county of residence. The
right panel is the same for in-migration, conditional on 1860 covariates of the individual’s 1939 county of residence.

slave and low-slave areas appear very similar to each other.
These results are specific to a specific five-year period in
American history for which we have data, but they do pro-
vide suggestive evidence that the differences we are seeing
are not due to geographic sorting alone.
In addition, we also provide additional evidence in ap-
pendix D, in which we analyze contemporary mobility data
from the 2000 US Census. These analyses show mobility
mostly from formerly low-slave to low-slave areas or from
formerly high-slave to high-slave areas (as opposed to mo-
bility between former low-slave and former high-slave areas,
or vice versa). Taken together, the evidence suggests that
geographic sorting is unlikely to be the exclusive explana-
tion behind our results.
Inequality and statistical discrimination
The final alternative explanation that we consider here is
that our regional patterns are the outcome of the contem-
porary economic landscape—in particular, contemporary
economic inequality between blacks and whites. O’Connell
(2012), for example, finds that slavery in 1860 predicts
black-white income inequality today (results echoed by
Lagerlöf [2005] and Nunn [2008]). Her results suggest that
our findings might be explained by a theory of statistical
discrimination (Becker 2010). In addition, because poverty
correlates with other traits (e.g., higher instances of crime)
then we might expect this type of statistical discrimination
to be even more pronounced. Consistent with O’Connell
(2012), we show in appendix table A.14 that slavery has an
effect on black-white income inequality measured using
both 1940 wages and 2014 median incomes, though the
effects have attenuated substantially since 1940. Could this
income inequality between blacks and whites in the South-
ern Black Belt be driving our findings on whites’ political
attitudes, for example, through statistical discrimination
(discrimination against blacks because they are, on average,
poorer than whites)?
We address this question with two analyses, both of
which cast doubt that income inequality and statistical
discrimination explain our results. First, we show that local
income inequality is a weak predictor of our main outcome
variables, and, if anything, the effects are in the opposite
direction as statistical discrimination theory would predict.
These are shown in table 4, which shows that inequality in
2014 is weakly related to political outcomes, and when it is
related, the relationships are in the opposite direction of
what would explain our results. In other words, counties
with higher black-white inequality are places where whites
are more likely to be Democrats, more likely to support
affirmative action, and are about as racially resentful as

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
632 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen

Table 4. Effect of Slavery versus the Effect of Inequality

Proportion Democrat Support for Affirmative Action Racial Resentment

(1) (2) (3) (4) (5) (6)

Log white-black income ratio, 2014 .040* .055** .0271 .041* 2.088 2.088
(.017) (.019) (.015) (.016) (.057) (.063)
Proportion slave, 1860 2.135** 2.158** .541**
(.044) (.037) (.144)
State fixed effects √ √ √ √ √ √
1860 covariates √ √ √ √ √ √
N 1,140 994 1,140 994 1,017 894
R2 .180 .226 .075 .124 .069 .159

Note. Inequality here is measured by the log of the ratio of white to black median incomes within counties in 2014. Data come from the American
Community Survey, 2009–14.
1
p ! .10.
* p ! .05.
** p ! .01.

whites elsewhere. Furthermore, the table also shows that
including contemporary black-white income inequality as a
covariate in our baseline specification does not substantially
change our estimates.19
Second, we check an observable implication of the in-
equality account. If income inequality drives slavery’s ef-
fects, then we would expect the marginal effect of slavery on
attitudes to vary in the income level of the respondent. For
example, higher-income respondents might be more (or
less) discriminatory on the basis of black-white income in-
equality. In appendix table A.15, we show that there is no
interaction between respondent income and proportion
slave. Taken together, these analyses provide evidence against
income inequality and economic conditions being the main
mechanism behind our findings.
AN EXPLANATION FOR SLAVERY’S EFFECT ON
CONTEMPORARY WHITE ATTITUDES BASED ON THE
HISTORICAL PERSISTENCE OF POLITICAL ATTITUDES
If the above explanations, largely rooted in contemporary
factors or population sorting, do not explain our findings,
then what does? In this section, we lay out the theory of
historical persistence in political attitudes. In our context,
this is the idea that slavery and its collapse reinforced re-
gional differences in political attitudes and these differences
have persisted over time.
The prevalence of slavery, coupled with the shock of its
removal, created strong incentives for Black Belt whites to
19. The coefficients on proportion slave in these models, however,
should be interpreted with caution due to the potential for post-treatment
bias.
try to preserve both their political and economic power by
promoting racially targeted violence, anti-black norms, and,
to the extent legally possible, racist institutions. These re-
inforced racial and political beliefs about black subjugation
within the Southern Black Belts, which, via institutional
path dependence and intergenerational socialization, have
persisted to the present day. While there is no question that
anti-black attitudes were rampant throughout the South
before the Civil War, we show in this section that even for
counties that were politically similar before the war, dif-
ferences in partisan voting became more pronounced in the
late nineteenth and early twentieth centuries. Thus, al-
though racism was prevalent across the high-slave and low-
slave South in the antebellum period, these areas diverge
greatly in terms of both institutionalized and socially en-
forced racism around the time of the Civil War, specifically
the period of Redemption that followed Reconstruction
(Foner 2011; Logan 1954).20
Below, we draw on work in both history and economics
to provide some qualitative background to the way in which
Southern white elites had incentives to foster and sustain
anti-black suppression in the former bastions of slavery.
We then provide evidence for our mechanism by pre-
senting evidence on (i) the burgeoning political importance
of race in the postbellum period, (ii) postbellum racial vi-
20. The more general idea behind this theory is that when an
entrenched social and economic institution like slavery is abruptly and
forcibly abolished, previously powerful groups (ex–slaveowning white
elite) seek to establish other local and informal institutions that serve a
similar purpose to that of the previous, forcibly abolished formal insti-
tution (slavery) (e.g., Acemoglu et al. 2011).

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

olence (including lynchings) and economic oppression,
(iii) the weakening of economic incentives for racial hos-
tility that took place as a result of the movement toward
agricultural mechanization beginning in the 1930s, and
(iv) evidence for intergenerational transmission of racial
attitudes.
Political and economic incentives for black
repression in the postbellum years
Why would racial attitudes in the Black Belt persist after
the Civil War, even as other regions of the country grad-
ually changed their views on race? Our explanation for this
postbellum divergence lies in the fact that, after emanci-
pation, Black Belt white elites faced two interrelated threats.
The first was political. In areas where blacks outnumbered
whites (in some cases 9 to 1), the abrupt enfranchisement
of blacks threatened white control over local politics (Du
Bois 1935; Key 1949; Kousser 1974). This gave whites in
former slaveholding counties an incentive to promote an
environment of violence and intimidation against the new
freedmen, with the purpose of election fraud and dis-
enfrachisement (Du Bois 1935; Kousser 1974). The second
threat to white elites was economic. The emancipation of
slaves after the Civil War was a major shock to the South-
ern economy: blacks now had to be paid wages (Higgs 1977).
Furthermore, emancipation brought blacks some freedom
over the amount of labor they supplied, and many ex-slaves
chose to work for themselves rather than for the white ruling
class (Ransom and Sutch 2001). This both reduced the labor
supply and increased labor costs sharply, threatening the
Southern plantation economy (Alston and Ferrie 1993; Ran-
som and Sutch 2001).21 Whites therefore had an incentive
to establish not just new forms of labor coercion that could
replace slavery but also new political restrictions that would
help protect white hegemony.
Since black populations and large-scale agricultural pro-
duction were greatest in former slaveholding counties, it was
in these counties that Southern elites exerted greater efforts
toward repression (Kousser 1974). These repressive tech-
niques are well documented in the economics and history
literatures (Alston and Ferrie 1993; Blackmon 2008; Lich-
tenstein 1996; Wiener 1978). For example, Wiener (1978,
62) describes how “planters used [Ku Klux] Klan terror to
keep blacks from leaving the plantation regions, to get them
to work, and keep them at work, in the cotton field.” Also
21. Some of these concerns were mitigated by the sharecropping sys-
tem that became pervasive in the postbellum period. This aligned the
incentives of the planters and the laborers to some extent (Ransom and
Sutch 2001, 88–89), but these arrangements did not reduce the incentives
of planters to further bolster their position in the labor market.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Volume 78 Number 3 July 2016 / 633
well documented is the fact that poor whites were complicit
with the landowning elite and would engage in and support
violent acts toward blacks, even though such violence could
presumably also lower white wages (Blackmon 2008; Du
Bois 1935; Roithmayr 2010). Note that these forces are dis-
tinct from the modern interpretation of racial threat, under
which whites would find high concentrations of black pop-
ulations undesirable. In our explanation, the Southern white
population sought to control, police, and dominate black
labor resources. Blacks were not threatening because of their
presence (as racial threat would imply); they were threat-
ening because they were an important provider of labor and,
in the post-emancipation environment, they could leave.
Three additional factors sustained and nurtured anti-
black attitudes in the former slaveholding counties, both
within the community and over time. The first is the con-
tinued and perhaps increased use of social norms that di-
rectly or indirectly put blacks in an inferior position relative
to whites (Du Bois 1935; Logan 1954).22 The second is the
expansion of these attitudes across local white communi-
ties, which could operate in a manner consistent with existing
theories of the diffusion of political opinions from elites to the
general public (Zaller 1992). The last is the intergenerational
transfer of political and racial attitudes, in a manner consistent
with theories of intergenerational socialization both in eco-
nomics and cultural anthropology (Bisin and Verdier 2000;
Boyd and Richerson 1988) as well as in political science
(Campbell et al. 1980; Jennings and Niemi 1968).23 In our
context, attitudes could be passed down from one generation
to the next through both cultural and institutional channels,
for example, via institutions such as Jim Crow or socially en-
forced segregation and racially motivated violence. In line with
the literature, such transmission would necessarily be imper-
fect, suggesting that there is some decay in these geographically
based relationships over time. This leads to one important
empirical prediction: the effects of slavery should be smaller in
22. The political and economic incentives for racial violence and op-
pression is likely to have produced or reinforced racially hostile attitudes
among whites through psychological and other channels. For example,
whites might have developed or maintained racially hostile attitudes to
minimize the “cognitive dissonance” associated with racially targeted vio-
lence toward blacks (Acharya, Blackwell, and Sen 2015). Theories in social
psychology, beginning with the work of Festinger (1957), would suggest that
engaging in violence could produce, or sustain, hostile attitudes among
members of the perpetrating group toward the victim group, if individuals
from the perpetrating group seek to minimize such dissonance.
23. We note that under our mechanism, it is sociopolitical attitudes,
rather than partisanship, that are passed down from parent to children.
This makes our mechanism consistent with the partisan realignment that
took place in the 1960s, given the assumption that partisanship depends at
least in some part on attitudes.
634 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
areas where the incentives for anti-black attitudes faded earlier.
Below, we provide evidence for this prediction.
Although our discussion focuses on the postbellum South,
our argument is not that racial hostility was nonexistent be-
fore the Civil War. Instead, the incentives that gained strength
at the time of emancipation likely exacerbated the political
differences between former slaveholding and non-slaveholding
areas. Racially hostile attitudes, in other words, may have
dissipated more quickly in areas that were non-slaveholding
and were less reliant on the provision of inexpensive black
labor. The South was surely racist before the Civil War, but
for this to explain our results, the strength of the racism
would have to align with the density of slavery. Our previous
analyses provide some limited evidence against this idea: in
results presented above, for example, our effects are robust to
controlling for the antebellum presence of freed blacks in the
county, which may be a possible indicator of antebellum ra-
cial attitudes. They are also robust to the inclusion of a host of
antebellum factors capturing economic and political differ-
ences. Below, we focus more closely on these ideas.
Timing of the political divergence and the
importance of the Reconstruction Period
What additional evidence do we have that racial attitudes
became politically salient in the postbellum period? To shed
light on this question, we examine the relationship between
proportion slave in 1860 and a long-standing historical mea-
sure of political attitudes, presidential vote shares. Until re-
alignment in the middle of the twentieth century, the Demo-
cratic Party was the racially conservative party, while the
Republican Party was the racially progressive party (Black
and Black 1987). In the period before the Civil War, the
Second Party system of Democrats and Whigs attempted
to remove slavery as a national issue, but slaveholding still
drove political differences on issues such as internal improve-
ments and tariffs, with elite slaveholders more likely to sup-
port Whig candidates (Holt 1999, 115–21). Do these differ-
ences drive the relationship between slavery and modern
attitudes? Or do post–Civil War events drive the results we
see above?
We address these questions by examining the effect of
slavery on county-level Democratic vote share of the pres-
idential vote between 1844 and 1968.24 To analyze the time
24. Since election outcomes are not disaggregated by voter race, these
data also include black voters. Blacks voted in large numbers following
emancipation but were systematically disenfranchised between in the late
nineteenth and early twentieth centuries. Large-scale re-enfranchisement
did not occur until the Voting Rights Act of 1965, which is why we stop
the analysis of such vote shares at that time.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
trend, we calculate the effect of a 25 percentage point in-
crease in 1860 slavery on the presidential vote in each year.
Thus, each point in figure 4 represents a (scaled) point es-
timate from a regression of county-level Democratic vote
share on county proportion slave in 1860, using the same
cotton-suitability IV design as table 2.25 As the figure shows,
there is little difference between slave and nonslave areas
before the Civil War, with the exception of 1856, where, if
anything, high-slave areas are more likely to vote for the
more moderate candidate on slavery, former president Mil-
lard Fillmore (compared to the relatively more pro-slavery
Democratic candidate James Buchanan). Thus, at least in
terms of national party politics, differences in white views
appear to organize around the density of slavery more strongly
after the Civil War.26
Instead, as figure 4 indicates, the data show that regional
differences emerged after the end of Reconstruction. As
white elites begin to restrict the vote of African Americans
in the late nineteenth century, the effect of slavery in-
creases, reaching its peak around the time that most of the
states finalized the enactment of poll taxes and literacy tests
to almost fully disenfranchise blacks (Kousser 1974). By
around 1950, the effect of slavery weakens, in part due to
some small additions of African Americans to the regis-
tered voter pool,27 and also to the move of national Dem-
ocratic candidates toward a civil rights platform. The effect
sizes from the 1950s and 1960s are roughly similar in
magnitude, but in the opposite direction, to the effect of
slavery on the Obama white vote in 2008, estimated from
CCES white respondents. Throughout, the difference in
voting behavior between formerly large slaveholding counties
and other counties is large and statistically significant. These
findings also hold if we use the WLS specification and con-
dition on our pretreatment covariates, and they hold even
when we control for county-level vote for Democrat James
Buchanan in the 1856 presidential election. These findings
suggest that large and significant differences in partisanship
emerge in the postbellum period even among counties that
were politically and economically similar in the antebellum
period. Furthermore, they reinforce the historical account of
Foner (2011, 11–18), who describes how the Civil War am-
25. We use the original 1860 county boundaries for the measurement
of slavery for outcomes until 1924, after which we use the data interpo-
lated to modern boundaries. The changes in county boundaries between
1924 and today in the sample states is minimal.
26. We obtain similar results for congressional elections and using the
WLS specification.
27. The percentage of the black voting-age population expanded from
3% to 18% in Georgia and from 0.8% to 13% in South Carolina (Mickey
2015).
 Volume 78 Number 3 July 2016 / 635

Figure 4. Effect of proportion slave on vote for Democratic presidential candidate in the South over time. Each point is the effect of a 25 percentage point
increase in proportion slave from separate IV models of county-level Democratic share of the presidential vote on proportion slave. Results for Obama in
2008 are from white respondents in the CCES.

plified divisions and resentments within the white Southern
population, concluding that “the [Civil War] redrew the
economic and political map of the white South” (17).
However, regarding antebellum differences, we note that
presidential election results in the prewar period could be
masking important intraregional differences, especially
since the Second Party system attempted to remove slavery
as an issue in national politics. For a stronger test of the
effect of slavery on antebellum politics, we turn to two
state-level elections that centered on slavery and sectional
issues. Specifically, after the Compromise of 1850, many
states in the South were divided over the deal’s fairness and
whether to extend popular sovereignty to the territories.
The deal, which admitted California as a free state, proved
so unpopular that several gubernatorial elections in 1851
pitted states’ rights extremists, who openly discussed se-
cession, against more moderate Unionist candidates, who
supported the compromise (Holt 1999, 605–9). In Georgia,
for example, Constitutional Union candidate Howell Cobb
ran against former Governor and Southern Rights candi-
date Charles McDonald, and, in Mississippi, moderate
Senator Henry Foote ran against former Senator (and fu-
ture Confederate President) Jefferson Davis. These elections
provided the (white) electorate a clear choice in the na-
tional fight over slavery and could potentially reveal ante-
bellum regional differences on the issue. Figure 5 presents
the relationship between slavery in 1850 and county-level
vote shares for the moderate candidates. In both states,
perhaps surprisingly, there is little evidence of a strong
relationship between slavery and vote choice, even in an
election that focused so heavily on the issue. Low-slave
areas were not actively anti-slavery and did not appear to
actively resist secession in the early 1850s. Furthermore, the
high-slave areas were at least as willing to compromise on
the issue of slavery, suggesting that the contemporary po-
litical geography of slavery probably does not trace its
origins to this time period.
Lynchings and other forms of suppression
Our incentives-based mechanism posits that violence was
used not only to disenfranchise blacks but also to suppress
their mobility and wages—a particularly strong incentive in
the post-war, post-emancipation period. If the mechanism
is at play, then we would expect to see greater racist vio-
lence in former slaveholding counties in this time period.
While we do not have measures of all forms of violent
racism in the post-Reconstruction Era, we do have county-

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
636 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen

Figure 5. Within-state relationship between proportion slave in 1850 in a county and percentage voting for Unionist candidates in 1851 in that county. Both
Howell Cobb and Henry Foote were running against strong States’ Rights supporters in the aftermath of the Compromise of 1850. Higher values on each
y-axis indicate more moderate voting with regard to slavery and secession.

level measures of one extreme form of racial violence:
lynchings (Beck and Tolnay 2004).28
In column 1 of table 5, we confirm the hypothesis that
the number of black lynchings between 1882 and 1930 per
100,000 1920 residents is greater in counties that had high
slave proportions in 1860, conditional on state-level fixed
effects and our 1860 covariates. The effect of slavery in 1860
and lynchings in this period is large and significant: a
10 percentage point increase in slave proportion is associ-
ated with a 1.89 increase in lynchings per 100,000 residents.
This result is in line with our incentive-based theory: there
is more racial violence in areas previously more reliant on
slave labor. Furthermore, if our explanation is correct, then
black farmers should be worse off in former slave areas due
to this greater local violence. Appendix table A.17 draws on
data from the the 1925 Agricultural Census (Haines 2010)
and shows that, in comparison to white farmers, black
farmers in former high-slave areas were significantly worse
off than those in other areas of the South. They were more
likely to be under tenancy agreements and less likely to own
their own farm.
Mechanization of Southern agriculture
After slaves were freed, Southern elites used various tactics
to gain an advantage in the market for black labor, in-
cluding racial violence, convict leasing, anti-vagrancy laws,
28. These data include all states in our analysis except Texas, Virginia,
and West Virginia.
and anti-enticement laws (Ransom and Sutch 2001). A
clear implication of our incentives-based mechanism is that
once the demand for black labor drops due to exogenous
technological development, the incentives for whites to
interfere in the labor market with such tactics should lessen,
and thus the effects of slavery on attitudes should also di-
minish. This implication is testable given that much of the
Southern economy was agricultural, and its main cash crop,
cotton, was heavily labor intensive until about the 1930s
when Southern agriculture started to mechanize and trac-
tors began to replace labor.
To test this, we examine whether counties that mecha-
nized earlier are indeed those where the effects of slavery
wanes more quickly. Following Hornbeck and Naidu
(2014), we use the number of farming tractors as a proxy
for mechanization.29 We interact the proportion of slaves
in 1860 with the change in the number of tractors per
100,000 acres of agricultural land in the county between 1930
and 1940, which we collect from the 1930 and 1940 Agri-
cultural Censuses (Haines 2010). We then estimate the effect
of this interaction on our three outcome measures. As table 5
shows, the effects of slavery are weaker for counties where
mechanization grew between 1930 and 1940. For example,
29. Tractors were an early form of mechanization for the cotton in-
dustry, intended to replace mule-drawn plows (Wiener 1978). Full-blown
mechanization of cotton production via the cotton picker would not be-
come widespread until after 1940.

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
 Volume 78 Number 3 July 2016 / 637

Table 5. Effect of Slavery on Postbellum Violence and Effect Modification by Mechanization

Lynchings Proportion Democrat Affirmative Action Racial Resentment

(1) (2) (3) (4)

Proportion slave, 1860 17.656** 2.183** 2.219** .703**

(5.661) (.049) (.042) (.161)
Tractors change, 1930–40 2.4171 2.629** 2.290**
(.243) (.207) (.791)
Proportion slave # tractors change 2.298** 2.226** 27.762**
(.868) (.741) (2.876)
State-year fixed effects √ √ √ √
1860 covariates √ √ √ √
N 910 1,145 1,145 1,020
R2 .350 .208 .104 .152
Specification WLS-county area WLS-sample size WLS-sample size WLS-sample size

Note. The first column is weighted least squares (WLS) with the total county area as weights (and where county area is omitted from the 1860 covariates,
though this has no effect on our analysis). The remaining columns are WLS with within-county sample size as weights. Lynchings are black lynchings
between 1882 and 1930 per 100,000 1920 residents (similar results hold using average population size between these dates). Tractors change is the change in
tractors per 100,000 acres of land between 1930 and 1940. Tractors in 1930 is the number of tractors per 100,000 acres of land in 1930.
1
p ! .10.
* p ! .05.
** p ! .01.

where mechanization did not grow between 1930 and 1940,
a 10 percentage point increase in proportion slave leads to a
1.8 percentage point drop in the percent of whites who
identify as Democrat today (95% confidence interval:
[22.7, 21.0]). Where mechanization grew rapidly, with
0.06 more tractors per 100,000 acres (90th percentile), the
same change in proportion slave leads to only a 0.2 per-
centage point decrease in the percent Democrat (95%
confidence interval: [21.1, 0.06]).
We note two potential concerns with this test. First,
the results could be consistent with a racial threat expla-
nation—early mechanization led to decreases in the black
population in these areas (as shown by Hornbeck and Naidu
2014). In appendix table A.16, however, we replicate the
analysis using the declines in proportion black from 1920 to
1940 and 1970 and find that areas with larger declines have,
if anything, larger effects associated with slavery. Thus, it
seems unlikely that dimishing racial threat is driving the at-
tenuating effects we see above. Second, it could be that more
racially tolerant counties chose to mechanize early in order to
rid themselves of the incentives for racial exploitation. How-
ever, as shown in appendix table A.18, there is no relationship
between growth in tractors and either racial violence or in-
equality in wages between blacks and whites. This casts doubt
that tractors or their growth are indicators of racial attitudes.
Moreover, as Hornbeck and Naidu (2014) argue, many of
the counties that mechanized early were those affected by
an exogenous shock, that of the Mississippi floods of 1927,
and are thus ex ante similar to counties that mechanized
later.
Intergenerational transmission of beliefs
Our last empirical analysis concerns the intergenerational
transfer of political attitudes—or, how attitudes on politics
and race have been passed down over time. One possibility
is that attitudes are shaped contemporaneously by local
institutions, for example, schools and churches, which have
themselves persisted. Another possibility is that these
attitudes have been passed down from parents to children,
perhaps within a broader context of institutions, but also
through intergenerational socialization. Both transmission
mechanisms are likely at play, and finding correlated
attitudes between parents and children would be consistent
with research in political science demonstrating the im-
portance of parents’ partisanship in shaping children’s
partisanship (e.g., Campbell et al. 1980; Jennings and Niemi
1968).
To explore the inheritance of racial attitudes from pre-
vious generations, we rely on the Youth-Parent Socializa-
tion Panel Survey, which measured the racial attitudes of a
national probability sample of high school senior students
in 1965 along with their parents (Elliot 2007; Jennings et al.
2005). This study followed up with the students and
interviewed them in 1973, 1982, and 1997, with response
rates over 80% in each survey wave. We focus on the subset
of this sample who lived in the South in 1965 (n p 241),

This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM

All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
638 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
which is a good proxy for students who were raised in the
South.30 If attitudes are being passed down from generation
to generation, then we would expect that the racial attitudes
of parents in 1965 should be correlated with the views of
their children in each wave of the survey. We measure
racial views in these surveys as the difference between the
thermometer score for whites and the thermometer score
for blacks, which is available for both parents and children
in each wave. Figure 6 shows these correlations for each
wave and demonstrates how stable this relationship is over
time. What these correlations show is that children with
racially conservative parents in 1965 are more likely to be
racially conservative themselves at least through age 50,
which is evidence of intergenerational socialization.
A final concern is that our findings regarding contem-
porary attitudes may be driven not by intergenerational
transmission but by the direct experiences and attitudes of
older individuals in our CCES sample. That is, we are
detecting an effect only among older whites, which would
suggest no or limited intergenerational transfer of attitudes
and instead an exposure to the tail end of Jim Crow. To test
this, we estimate the effect among a subgroup who were more
likely to receive such attitudes only from their parents: whites
born after the Civil Rights Act of 1964 and the Voting Rights
Act of 1965. While segregation and race-related oppression
extended well past this time, both pieces of legislation have
been acknowledged as influential in strongly reducing segre-
gation in the South and increasing black enfranchisement
(Rosenberg 2008). In table 6, we show the effect of slavery on
these younger whites is just as strong and statistically signifi-
cant as it is for older whites. Together with our results on racial
threat and income-based discrimination, these findings pro-
vide some evidence that the parent-to-child transmission
mechanism is an important component of how slavery affects
attitudes. As such, this evidence provides some support that
political culture, rather than exclusively institutions, plays an
important role in explaining the persistent effects of American
slavery.
CONCLUSION
In this paper, we have shown that an institution that was
formally abolished 150 years ago still has effects on attitudes
today. Specifically, we show that American slavery has had a
direct impact on Southern whites’ (i) partisan identification,
(ii) attitudes on affirmative action, (iii) levels of racial re-
sentment, and (iv) attitudes toward blacks. We further
30. See Jennings, Stoker, and Bowers (2009), for a fuller analysis of the
national results from this panel survey.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
showed that our findings are robust to instrumenting for
cotton suitability, lending credibility that our estimates are
causal. Our findings are robust to accounting for a wide
variety of factors that could plausibly affect both the share
of the population that was enslaved and also contempo-
rary political attitudes. In addition, we ruled out several
contemporary-based explanations. Specifically, we ruled out
the theory of “racial threat,” or the idea that contemporary
shares of the black (minority) population are what drive
white (majority) group racial attitudes. When we took into
account contemporary shares of the black population using
appropriate methods, we found that slavery continues to
have a separate direct effect. We also provided suggestive
evidence that our results are not due exclusively to geo-
graphic mobility over the course of the twentieth century,
nor are they due to contemporary income inequality be-
tween blacks and whites.
Our results instead suggest a separate causal channel
attributable to the historical persistence of regional varia-
tion in attitudes that was amplified by post–Civil War events.
The years during and after the Reconstruction Period saw
whites coordinating to provide an informal social infra-
structure (and to the extent legally permissible an institu-
tional one as well) to maintain as much as possible the
economic and political power previously guaranteed to them
under slavery. As affirmative support, we showed that greater
prevalence of slavery predicts more conservative (for many
years more Democratic) presidential vote shares, higher rates
of radical violence, and decreased wealth concentrated in
black farms in the decades after Reconstruction. We also
showed that the long-term effects of slavery are smaller in
Figure 6. Correlation coefficients between parent’s white racial preference
in 1965 and their children’s white racial preference measured at four time
periods: 1965, 1973, 1982, and 1997. White racial preference here is
measured as the difference between the respondent’s thermometer scores
for white and black people.

Table 6. Effects of Slavery for Those Born after the Voting Rights Act
Proportion Democrat Logistic
(1)
Proportion slave, 1860 21.113**
(.365)
State-year fixed effects √ √
1860 Covariates √ √
N 9,551
R2
Note. All models are at the individual level, with standard errors, in parentheses, clustered on county and weighted by CCES survey weights.
* p ! .05.
** p ! .01.
areas of the US South that were quick to mechanize in the
early to mid-twentieth century. Finally, we also offered evi-
dence that parent-to-child transmission could be an impor-
tant mechanism by which attitudes have been passed down
over time. However, we do not rule out that Southern insti-
tutions may have also played an important role.
Our research has substantial implications for our un-
derstanding and study of politics. Much work within po-
litical science, and on public opinion specifically, focuses
on contemporary respondent factors, such as income, age,
gender, education, and so on, in trying to explain contem-
porary public opinion. However, this style of research over-
looks historical and culturally rooted explanations for the
formation of public opinion. Our findings here suggest that
historical institutions like slavery are significant in shaping
American culture and politics, even if they no longer exist.
In light of this, the field of political behavior could benefit
from exploring other potential relationships between his-
torical forces and contemporary attitudes. As Key (1949, 6)
himself observed, social and historical forces have “an im-
pact on political habit whose influence has not worn away
even yet.” This might be the case not just for development
of political attitudes in the US South but also in other arenas
within American politics and elsewhere in the world.
ACKNOWLEDGMENTS
We thank Stephen Ansolabehere, Larry Bartels, David Broock-
man, Cathy Cohen, Stanley Engerman, Gerald Gamm, Hein
Goemans, Justin Grimmer, Steven Hahn, Jeff Jenkins, Gary
King, George Krause, Chyrl Laird, Corrine McConnaughy,
Marc Meredith, Robert Mickey, Clayton Nall, Suresh Naidu,
Dianne Pinderhughes, Kevin Quinn, Karthick Ramakrish-
nan, Marc Ratkovic, John Roemer, Daria Roithmayr, Cyrus
Samii, Eric Schickler, Ken Shotts, Adam Slez, Paul Sniderman,
Brandon Stewart, Michael Tesler, Vesla Weaver, Jon Woon,
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Volume 78 Number 3 July 2016 / 639
Affirmative Action Logistic Racial Resentment OLS
(2) (3)
2.840* .672*
(.358) (.265)
√
√
9,528 3,350
.059
and seminar participants at Boston University, Columbia,
Dartmouth, Harvard Kennedy School, Harvard Law School,
Notre Dame, Princeton, Rochester, Stanford, Warwick,
University of California, Berkeley, University of California,
Riverside, University of Pennsylvania, University of Pitts-
burg, Washington University–St. Louis, University of South-
ern California, and University of Virginia for comments and
suggestions. We are also grateful to Michael Haines, Eitan
Hersh, and Heather O’Connell for sharing data with us.
REFERENCES
Acemoglu, Daron, Davide Cantoni, Simon Johnson, and James A.
Robinson. 2011. “The Consequences of Radical Reform: The French
Revolution.” American Economic Review 101 (7): 3286–3307.
Acemoglu, Daron, Camilo García-Jimeno, and James A. Robinson. 2012.
“Finding Eldorado: Slavery and Long-Run Development in Colombia.”
Journal of Comparative Economics 40 (4): 534–64.
Acharya, Avidit, Matthew Blackwell, and Maya Sen. 2015. “Explaining
Attitudes from Behavior: A Cognitive Dissonance Approach.” HKS
Faculty Research Working paper RWP14–026, Harvard Kennedy
School, Harvard University.
Acharya, Avidit, Matthew Blackwell, and Maya Sen. Forthcoming.
“Explaining Causal Findings without Bias: Detecting and Assessing
Direct Effects.” American Political Science Review. http://www.matt
blackwell.org/files/papers/direct-effects.pdf.
Alston, Lee J., and Joseph P. Ferrie. 1993. “Paternalism in Agricultural
Labor Contracts in the U.S. South: Implications for the Growth of the
Welfare State.” American Economic Review 83 (4): 852–76.
Ansolabehere, Stephen. 2010. “Cooperative Congressional Election Study,
2010: Common Content.” Harvard University [producer]. http://projects
.iq.harvarfd,edu/cces/.
Beck, E. M., and Stewart E. Tolnay. 2004. “Confirmed Inventory of
Southern Lynch Victims, 1882–1930.” Data file available from authors.
Becker, Gary S. 2010. The Economics of Discrimination. Chicago: Uni-
versity of Chicago Press.
Bisin, Alberto, and Thierry Verdier. 2000. “A Model of Cultural Trans-
mission, Voting and Political Ideology.” European Journal of Political
Economy 16 (1): 5–29.
Black, Earl, and Merle Black. 1987. Politics and Society in the South.
Cambridge, MA: Harvard University Press.
640 / The Political Legacy of American Slavery Avidit Acharya, Matthew Blackwell, and Maya Sen
Blackmon, Douglas A. 2008. Slavery by Another Name: The Re-enslavement
of Black Americans from the Civil War to World War II. New York:
Anchor Books.
Blalock, Hubert M. 1967. Toward a Theory of Minority-Group Relations.
Hoboken, NJ: Wiley.
Blumer, Herbert. 1958. “Race Prejudice as a Sense of Group Position.”
Pacific Sociological Review 1 (1): 3–7.
Boyd, Robert, and Peter J. Richerson. 1988. Culture and the Evolutionary
Process. Chicago: University of Chicago Press.
Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes.
1980. The American Voter. Chicago: University of Chicago Press.
Carmines, Edward G., and James A. Stimson. 1989. Issue Evolution: Race
and the Transformation of American Politics. Princeton, NJ: Princeton
University Press.
Dell, Melissa. 2010. “The Persistent Effects of Peru’s Mining Mita.”
Econometrica 78 (6): 1863–1903.
Du Bois, W. E. B. 1935. Black Reconstruction in America, 1860–1880. New
York: Free Press.
Elliot, Patrick. 2007. “Three Generations Combined, 1965–1997.” Inter-
university Consortium for Political and Social Research (ICPSR).
http://doi.org/10.3886/ICPSR04532.v1.
Faust, Drew Gilpin. 1988. The Creation of Confederate Nationalism: Ide-
ology and Identity in the Civil War South. Baton Rouge: Louisiana
State University Press.
Festinger, Leon. 1957. A Theory of Cognitive Dissonance. Stanford, CA:
Stanford University Press.
Foner, Eric. 2011. Reconstruction: America’s Unfinished Revolution, 1863–
1877. New York: Harper & Row.
Gilens, Martin. 2009. Why Americans Hate Welfare: Race, Media, and the
Politics of Antipoverty Policy. Chicago: University of Chicago Press.
Giles, Micheal W. 1977. “Percent Black and Racial Hostility: An Old
Assumption Reexamined.” Social Science Quarterly 58 (3): 412–17.
Giles, Micheal W., and Melanie A. Buckner. 1993. “David Duke and Black
Threat: An Old Hypothesis Revisited.” Journal of Politics 55 (3): 702–13.
Glaser, James M. 1994. “Back to the Black Belt: Racial Environment and
White Racial Attitudes in the South.” Journal of Politics 56 (1): 21–41.
Green, Donald P., and Lynn Vavreck. 2008. “Analysis of Cluster-Randomized
Experiments: A Comparison of Alternative Estimation Approaches.” Po-
litical Analysis 16 (2): 138–52. http://pan.oxfordjournals.org/content/16/2
/138.abstract.
Haines, Michael R. 2010. “Historical, Demographic, Economic, and Social
Data: The United States, 1790–2002.” Inter-university Consortium for
Political and Social Research (ICPSR) [distributor]. http://www.icpsr
.umich.edu/icpsrweb/ICPSR/studies/2896.
Higgs, Robert. 1977. Competition and Coercion: Blacks in the American
Economy, 1865–1914. Cambridge: Cambridge University Press.
Holt, Michael F. 1999. The Rise and Fall of the American Whig Party:
Jacksonian Politics and the Onset of the Civil War. Oxford: Oxford
University Press.
Hopkins, Daniel J. 2010. “Politicized Places: Explaining Where and When
Immigrants Provoke Local Opposition.” American Political Science
Review 104 (1): 40–60.
Hornbeck, Richard, and Suresh Naidu. 2014. “When the Levee Breaks:
Black Migration and Economic Development in the American South.”
American Economic Review 104 (3): 963–90.
Imai, Kosuke, Luke Keele, Dustin Tingley, and Teppei Yamamoto. 2011.
“Unpacking the Black Box of Causality: Learning about Causal
Mechanisms from Experimental and Observational Studies.” Ameri-
can Political Science Review 105 (4): 765–89.
Jennings, M. Kent, Gregory B. Markus, Richard G. Niemi, and Laura
Stoker. 2005. “Youth-Parent Socialization Panel Study, 1965–1997:
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Four Waves Combined.” Inter-university Consortium for Political and
Social Research (ICPSR). Data set. http://doi.org/10.3886/ICPSR04037.v1.
Jennings, M. Kent, and Richard G. Niemi. 1968. “The Transmission of
Political Values from Parent to Child.” American Political Science
Review 62 (1): 169–84.
Jennings, M. Kent, Laura Stoker, and Jake Bowers. 2009. “Politics across
Generations: Family Transmission Reexamined.” Journal of Politics 71
(3): 782–99.
Katznelson, Ira, Kim Geiger, and Daniel Kryder. 1993. “Limiting
Lliberalism: The Southern Veto in Congress, 1933–1950.” Political
Science Quarterly 108 (2): 283–306.
Key, V. O. 1949. Southern Politics in State and Nation. New York: Knopf.
Kinder, Donald R., and David O. Sears. 1981. “Prejudice and Politics:
Symbolic Racism versus Racial Threats to the Good Life.” Journal of
Personality and Social Psychology 40 (3): 414–31.
Kousser, Morgan J. 1974. The Shaping of Southern Politics: Suffrage Re-
striction and the Establishment of the One-Party South, 1880–1910.
New Haven, CT: Yale University Press.
Kuziemko, Ilyana, and Ebonya Washington. 2015. “Why Did the
Democrats Lose the South? Bringing New Data to an Old Debate.”
Working paper no. 21703, National Bureau of Economic Research,
Cambridge, MA.
Lagerlöf, Nils-Petter. 2005. “Geography, Institutions, and Growth: The
United States as a Microcosm.” Working paper, Department of Eco-
nomics, York University, Toronto.
Lee, Woojin, and John E. Roemer. 2006. “Racism and Redistribution in the
United States: A Solution to the Problem of American Exceptional-
ism.” Journal of Public Economics 90 (6): 1027–52.
Lichtenstein, Alex. 1996. Twice the Work of Free Labor: The Political
Economy of Convict Labor in the New South. Brooklyn, NY: Verso.
Logan, Rayford W. 1954. The Negro in American Life and Thought. New
York: Dial.
Markus, Gregory B., and Philip E. Converse. 1979. “A Dynamic Simul-
taneous Equation Model of Electoral Choice.” American Political
Science Review 73 (4): 1055–70.
Mickey, Robert. 2015. Paths out of Dixie: The Democratization of Au-
thoritarian Enclaves in America’s Deep South, 1944–1972. Princeton,
NJ: Princeton University Press.
Mitchener, Kris James, and Ian W. McLean. 2003. “The Productivity of
U.S. States since 1880.” Journal of Economic Growth 8 (1): 73–114.
Nunn, Nathan. 2008. “Slavery, Inequality, and Economic Development in
the Americas.” In Elhanan Helpman, ed., Institutions and Economic
Performance. Cambridge, MA: Harvard University Press.
Nunn, Nathan. 2009. “The Importance of History for Economic Devel-
opment.” Annual Review of Economics 1 (1): 65–92.
Nunn, Nathan, and Leonard Wantchekon. 2011. “The Slave Trade and the
Origins of Mistrust in Africa.” American Economic Review 101 (7):
3221–52.
O’Connell, Heather A. 2012. “The Impact of Slavery on Racial Inequality
in Poverty in the Contemporary U.S. South.” Social Forces 90 (3): 713–
34.
Ogburn, William F., and Charles M. Grigg. 1956. “Factors Related to the
Virginia Vote on Segregation.” Social Forces 34 (4): 301–8.
Oliver, J. Eric, and Tali Mendelberg. 2000. “Reconsidering the Environ-
mental Determinants of White Racial Attitudes.” American Journal of
Political Science 44:574–89.
Pierson, Paul. 2000. “Increasing Returns, Path Dependence, and the Study
of Politics.” American Political Science Review 94 (2): 251–67.
Ransom, Roger L., and Richard Sutch. 2001. One Kind of Freedom: The Eco-
nomic Consequences of Emancipation. Cambridge: Cambridge University
Press.

Reed, John Shelton. 1972. “Percent Black and Lynching: A Test of
Blalock’s Theory.” Social Forces 50 (3): 356–60.
Roithmayr, Daria. 2010. “Racial Cartels.” Michigan Journal of Race and
Law 16 (1): 45–79.
Rosenbaum, Paul R. 1984. “The Consquences of Adjustment for a Con-
comitant Variable That Has Been Affected by the Treatment.” Journal
of the Royal Statistical Society: Series A (General) 147 (5): 656–66.
Rosenberg, Gerald N. 2008. The Hollow Hope: Can Courts Bring about
Social Change? Chicago: University of Chicago Press.
Ruggles, Steven, J. Trent Alexander, Katie Genadek, Ronald Goeken,
Matthew B. Schroeder, and Matthew Sobek. 2010. “Integrated Public
Use Microdata Series: Version 5.0” [Machine-readable database].
Valentino, Nicholas A., and David O. Sears. 2005. “Old Times There
Are Not Forgotten: Race and Partisan Realignment in the Contem-
porary South.” American Journal of Political Science 49 (3): 672–
88.
Vansteelandt, Sijn. 2009. “Estimating Direct Effects in Cohort and Case–
Control Studies.” Epidemiology 20 (6): 851–60.
This content downloaded from 128.103.149.052 on June 15, 2016 06:28:11 AM
All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).
Volume 78 Number 3 July 2016 / 641
Voigtländer, Nico, and Hans-Joachim Voth. 2012. “Persecution Perpetu-
ated: The Medieval Origins of Anti-Semitic Violence in Nazi Germany.”
Quarterly Journal of Economics 127 (3): 1339–92.
Voss, D. Stephen. 1996. “Beyond Racial Threat: Failure of an Old Hy-
pothesis in the New South.” Journal of Politics 58 (4): 1156–70.
Wiener, Jonathan M. 1978. Social Origins of the New South: Alabama,
1860–1885. Baton Rouge: Louisiana State University Press.
Wilcox, Clyde, Lee Sigelman, and Elizabeth Cook. 1989. “Some Like It Hot:
Individual Differences in Responses to Group Feeling Thermometers.”
Public Opinion Quarterly 53 (2): 246–57.
Woodward, C. Vann. (1955) 2002. The Strange Career of Jim Crow. Ox-
ford: Oxford University Press.
Wright, Gavin. 1986. Old South, New South: Revolutions in the Southern
Economy Since the Civil War. Vol. 2. New York: Basic Books.
Wright, Gerald C., Jr. 1977. “Contextual Models of Electoral Behavior: The
Southern Wallace Vote.” American Political Science Review 71 (2): 497–508.
Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge:
Cambridge University Press.